Arxiv:2007.00015V2 [Hep-Ph] 15 Aug 2020 Terparts

FERMILAB-PUB-20-257-T

Tau neutrinos at DUNE: new strategies, new opportunities

Pedro Machado,1 Holger Schulz,2 and Jessica Turner1 1Theoretical Physics Department, Fermi National Accelerator Laboratory, P.O. Box 500, Batavia, IL 60510, USA. 2Department of Physics, University of Cincinnati, Cincinnati, OH 45219, USA. (Dated: August 18, 2020) We propose a novel analysis strategy, that leverages the unique capabilities of the DUNE experiment, to study tau neutrinos. We integrate collider physics ideas, such as jet clustering algorithms in combination with machine learning techniques, into neutrino measurements. Through the con- struction of a set of observables and kinematic cuts, we obtain a superior discrimination of the signal (S) over the√ background (B). In a single year, using the nominal neutrino beam mode, DUNE may achieve S/ B of 3.3 and 2.3 for the hadronic and leptonic decay√ channels of the tau respectively. Operating in the tau-optimized beam mode would increase S/ B to 8.8 and 11 for each of these channels. We premier the use of the analysis software Rivet, a tool ubiquitously used by the LHC experiments, in neutrino physics. For wider accessibility, we provide our analysis code.

I. INTRODUCTION near-to-far detector ratios, will be crucial in limiting systematic uncertainties in future studies of tau neutrinos Arguably, the tau neutrino is the least understood at long-baseline neutrino experiments. particle of the Standard Model. Thus far a total of The observation of tau neutrinos in current neutrino 14 tau neutrinos have been positively identified by the beams relies on tau appearance due to oscillations. The DONuT [1] and OPERA [2] experiments. The former de- phase of the oscillation is given by terms such as tected beam tau neutrinos from the decay of D mesons. s ∆m2L ∆m2/eV2L/km The latter observed, for the first time, νµ → ντ oscilla- sin2 ' sin2 1.27 , (1) tions. In both experiments, the identification of τ leptons 4E Eν /GeV

produced by ντ charged current (CC) interactions relies 2 upon the reconstruction of characteristic event topolo- where ∆m is a mass squared splitting, Eν is the neutrino energy and L is the experimental baseline. The gies: the τ lepton leaves a millimeter-scale track in the 2 larger of the two mass squared splittings is ∆matm ' detector emulsion followed by a kink from its subsequent −3 2 decay. In addition to DONuT and OPERA, the pres- 2.5 × 10 eV and the neutrino energy threshold to cre- ence of tau neutrinos has been statistically inferred from ate a tau lepton from CC interactions is Eν & 3.5 GeV. Therefore, the baseline necessary to maximize ντ ap- νµ → ντ oscillations of multi-GeV atmospheric neutrinos by Super-Kamiokande [3, 4] and IceCube [5]. These pearance is of the order 2000 − 3000 km. At such long distances, accelerator neutrino experiments require ex- searches are based on the ντ contribution to the number of hadronic and/or leptonic neutrino events. tremely powerful neutrino sources to amass sufficient Despite the excellent reconstruction capabilities of statistics in order to study the ντ sector. A further diffi- DONuT and OPERA and the large statistics of Super- culty associated with ντ detection is that the decay of a τ Kamiokande and IceCube, tau neutrino observables, such lepton always includes a ντ in the final state which carries as ν -nuclei cross sections and oscillation parameters ex- away a fraction of undetectable energy. Therefore, recon- τ struction of the original beam neutrino energy is a chal- tracted from ντ measurements, have large statistical and systematic uncertainties. Indeed, the tau neutrino nu- lenging task. Although atmospheric neutrinos have ener- cleon interaction cross section has larger uncertainties gies well above the tau production threshold and travel [1, 4–6] than its electronic [7–9] and muonic [10–12] coun- the necessary distances to induce large νµ → ντ oscillations, reconstructing the energy of these neutrinos is arXiv:2007.00015v2 [hep-ph] 15 Aug 2020 terparts. Super-Kamiokande provided the most accurate challenging as the direction of the incident particle is not measurement of the ντ CC cross section using atmospheric neutrinos and this has an uncertainty of 21% [4]. known on an event-by-event basis. Moreover, the unoscillated tau neutrino flux itself is a Despite such difficulties, upcoming multi-purpose neu- source of systematic uncertainty as it depends on the trino experiments, such as DUNE [14], are well posed to detect tau neutrinos given their large fiducial volume, Ds production rate; the accuracy of which is limited by the incomplete understanding of hadronic effects. These powerful neutrino beam and exquisite track reconstruc- two sources of uncertainty can be mitigated via the di- tion capability. There have been a number of works which rect study of tau neutrino production as proposed by the explore the tau neutrino sector including pioneering pro- DsTau experiment [13]. This collaboration aims to pro- posals for observing ντ at beam dump experiments [15] and the subsequent experimental search by NOMAD [16], vide an independent ντ flux prediction for future neutrino beams with an uncertainty below 10%. As such, the sys- polarization effects on τ decay products for atmospheric and beam ντ searches [17–21], optimization of cuts on τ tematic error of the CC ντ cross section prediction can be lowered. This measurement, together with the use of decay products in the DUNE beam neutrino sample [22], tests of unitarity and patterns of leptonic mixing [23–25] 2

1011 and beyond the Standard Model physics probes with ντ νe measurements [26–29]. νµ In this paper, we are particularly interested in future ντ liquid argon time projection chamber (LArTPC) experi- νe τ-optimized ments such as DUNE which have proven to be endowed νµ τ-optimized with excellent event topology reconstruction capabilities. 1010 ντ τ-optimized The future DUNE experiment will combine bubble cham-

ber quality data with calorimetry and large statistics. It year ν 2

will therefore provide an unprecedented opportunity to m study the ντ sector. We begin in Section II with a general discussion of the DUNE experiment and tau neutrino 109 detection at LArTPCs. We perform a sophisticated simulation of the tau neutrino signal and background processes in DUNE, taking into account tau lepton polarization and nuclear physics effects. This is outlined in 108 Section III where we build on and expand the analysis 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0 performed in Ref. [22] by using modern techniques, such Eν [GeV] as jet clustering algorithms and deep neural networks, in order to optimize signal-to-background ratios. Further- FIG. 1. Nominal (solid) and tau-optimized (dashed) neutrino more, we quantify the importance of charge identification fluxes at the DUNE far detector neutrino for νe (blue), νµ of pions and running in the tau-optimized beam configu- (green) and ντ (red). ration for tau neutrino searches. Finally, we summarize and discuss future studies, which will incorporate detector effects, in Section IV. will be identified and reconstructed with 100% efficiency in the detector [32] as detailed in Section III. We ac- knowledge our approach is optimistic as we do not apply II. TAU NEUTRINOS AT DUNE detector effects. Moreover, application of such effects will inevitably deteriorate the signal-to-background ra- Before discussing tau neutrino events in detail, we first tio; however, the purpose of this paper is to establish a provide a description of the key aspects of the DUNE ex- new tool chain and apply novel techniques which will be periment. The neutrino beam at DUNE is produced by of use in a more detailed future study. a 120 GeV proton beam hitting a target. The nominal Conservatively, we assume neutrons are completely in- beam power will be approximately 1.2 MW and is ex- visible at LArTPCs. However, the techniques imple- 21 pected to deliver 1.1 × 10 protons on target (POT) per mented in this work could benefit from information on year. The far detector consists of a 40 kiloton fiducial the energy deposited by secondary hard neutron-proton mass LArTPC with a baseline of 1300 km. scattering or multiple neutron scatterings. Such pro- In Fig. 1 we show the oscillated neutrino fluxes at the cesses can constitute a considerable fraction of the total DUNE far detector for the nominal neutrino mode (solid) hadronic energy of an event [33]. Exploring this possi- and tau-optimized configuration (dashed) for each neu- bility would require a fully-fledged detector simulation trino flavor [30]. The oscillation parameters chosen as which is beyond the scope of this paper. inputs throughout this paper are taken from global fit Finally, we exploit DUNE’s capability to identify the data [31]: charge of pions via the topology of the pion tracks (see

2 −5 2 2 −3 2 e.g. Refs. [32, 34, 35]). Charged pions can undergo ∆m21 = 7.4 × 10 eV , ∆m31 = 2.5 × 10 eV , many different processes as they travel through a dense 2 2 2 s12 = 0.31, s13 = 0.0224, s23 = 0.5, δCP = 1.2π, medium. These processes include two-body pion absorption (π+np → pp, π−np → nn), elastic scattering, single + 0 − 0 where sij ≡ sin θij. In the nominal neutrino mode, the charge exchange (π n → π p, π p → π n) and inelas- fluxes of all three neutrino flavors peak in the range 1-3 tic scattering. Most importantly, stopped π− are typi- GeV. Therefore many of the tau neutrinos (and antineu- cally captured by the positively charged argon nucleus trino contaminants) have energies below the tau lepton as opposed to stopped π+ which simply decay to µ+ fol- production threshold. While the integrated flux of the lowed by a Michel e+. Due to these distinctively dif- tau-optimized mode is similar to the nominal mode, the ferent topologies, charge identification of pions is possi- peak of the spectra for all three flavors is broader and ble on an event-by-event basis. It should be noted there consequently there are more tau neutrinos with energies have also been discussions on exploiting topological infor- above the tau production threshold at the far detector. mation [36, 37] to statistically distinguish the charge of In this study, we do not consider µ/π misidentifica- muons, see e.g. Section 5.5.2.1 of Ref. [32]. In principle, tion or energy and angular resolutions. Furthermore, we similar techniques could be used to infer charge identi- assume that particles above a certain energy threshold fication of pions statistically [38]. We study the impact 3

ν Decay mode Branching ratio τjet τ e,µ Leptonic 35.2% ντ τ ντ τ ν , ν − e µ e ν¯eντ 17.8% − µ ν¯µντ 17.4% Hadronic 64.8% π−π0ν 25.5% τ n n − njet njet π ντ 10.8% − 0 0 π π π ντ 9.3% − − + π π π ντ 9.0% − − + 0 π π π π ντ 4.5% ν ν νe,νµ e,µ other 5.7% α α

TABLE I. Dominant decay modes of τ −. All decays involving kaons, as well as other subdominant decays, are in the “other” category. n njet n njet of perfect pion charge identification on signal and back- + FIG. 2. Pictorial representation of hadronic tau (upper left) ground yields and contrast this with the case π cannot and leptonic tau (upper right) signals, and their correspond- − be distinguished from π . ing backgrounds (lower). Now that we have discussed the details of the DUNE experiment and the assumptions underlying our analysis, we proceed onto the physics of taus and tau neutrinos. in this channel in Section III B. As the DUNE beam is predominantly comprised of muon Finally, the dominant background to τhad are the neu- neutrinos, the main sample of tau neutrinos at the far de- tral current (NC) neutrino scattering events which have tector are due to νµ → ντ oscillations. At the oscillation contributions from all three neutrino flavors. Despite the minimum of Eν ∼ 3 GeV, the majority of muon neutrinos fact that all neutrino flavors contribute to NC events (in- are expected to have oscillated to tau neutrinos. cluding tau neutrinos), the NC cross section is smaller A key element in the study of tau neutrino physics than the CC cross section [39, 40]. Furthermore, the is the decay modes of the tau lepton. The most relevant hadronic branching fraction of taus is almost twice as tau branching ratios are given in Table I. In the following, large as the leptonic branching fraction (see Table I). we denote the sample of taus that decay to electrons and Consequently, the hadronic decays of the tau have a muons as τe and τµ respectively while hadronically decay- higher signal-to-background ratio, in the nominal beam ing taus will be denoted as τhad. The tau decay length mode, than either of the leptonic channels as we outline has a value of cτ ≈ 87 µm, which is much larger than the in Section III A. We note that there is a small contribu- argon nuclear radius (of about 3.4 fm), and thus the tau tion to the signal and background from ντ CC and νe,µ,τ decays far outside the nucleus. Subsequently, its decay NC events respectively which we include in our analysis. products are not subject to the argon’s nuclear potential. A pictorial summary of the dominant tau signals and However, the tau lifetime is too short to lead to observ- backgrounds is shown in Fig. 2. In the upper right (left) able displaced vertices in DUNE where the granularity we show the leptonic (hadronic) decay of the tau and in is limited by the typical wire spacing of a few millime- the lower right (left) its associated dominant background. ters. It is thus unlikely that tau tracks can be observed The target nucleon inside the argon nucleus is denoted 1 at DUNE . This, together with the severe background, as n and the green cones represent the (mostly) hadronic makes ντ detection particularly challenging. activity that emerges from the argon nucleus after the The background of the τµ signal stems mainly from hadronization and subsequent intra-nuclear cascade. We 2 νµ CC events . This channel is widely considered to be denote this collection of particles, emerging from the nu- experimentally intractable as the νµ flux is prohibitively cleus, as njet because we apply a jet clustering algorithm large. Similarly, the dominant background of the τe sig- to the signal and background events. Likewise, in the nal are νe CC events. As the νe flux at DUNE is a small case of the hadronic decays of the tau we represent the fraction of the total neutrino flux, we study ντ detection final states as τjet.

III. SIMULATION AND RESULTS 1 It is possible a handful of tau tracks could be observed from atmospheric tau neutrinos as they have very high energies. 2 There is a subdominant contribution from the νµ CC interac- In order to account for nuclear physics eﬀects, the tions. signal and background neutrino-nucleon interactions are 4

i 10 3.5 N h [GeV]

i 3 8 E ∑ h 2.5

6 2

4 1.5

signal 1 signal 2 νµ BG νµ BG τ-optimized signal 0.5 τ-optimized signal τ-optimized νµ BG τ-optimized νµ BG 0 0 0 5 10 15 20 25 30 0 5 10 15 20 25 30 Eν [GeV] Eν [GeV]

FIG. 3. The left (right) plot shows the the mean number (average sum of energies) of visible particles as a function of the true neutrino energy Eν for the hadronic channel. The signal from the nominal (tau optimized) beam mode is shown in solid (dashed) red and its dominant background in solid (dashed) blue. simulated using the GiBUU event generator [41]. As dis- Jet clustering algorithms are an essential tool for a vari- cussed before, the signal process is a CC interaction of ety of LHC studies however we demonstrate their utility a tau neutrino or antineutrino with the argon nucleus. for both the τhad and τe channels. We undertake this This interaction produces a tau which then decays far treatment for two reasons: first, the physics of individual outside the argon nucleus. The Monte-Carlo events out- meson formation depends on the hadronization process, put by GiBUU factorize into the stable, polarized tau which is largely incalculable, while jets are objects con- and other final state particles such as pions, protons structed to capture the underlying hard physics which is and neutrons. The latter are products of the propaga- much better understood. Second, applying a jet cluster- tion or recoiled/created nucleons throughout the nuclear ing algorithm via FastJet [44] is straightforward in Rivet medium and are thus subject to the nuclear potential, and we demonstrate this technique to be a useful method re-scattering and absorption processes. of characterizing event topologies at LArTPCs. The tau lepton produced by a neutrino CC interac- We divide our analysis into the hadronic and lep- tion will be polarized and the distributions of its decay tonic channels and present them in Section III A and products will critically depend on its spin polarization. Section III B respectively. We have used CP -optimized Therefore it is important to consider the spin polariza- fluxes derived for the forward horn current polarity (neu- tion of taus in addition to their production cross sec- trino mode) unless otherwise specified. We will also tions. This has been discussed at length for tau neu- present DUNE’s sensitivity to tau neutrinos in the tau- trinos at OPERA [17] and also atmospheric neutrinos optimized beam configuration. at Super-Kamiokande [20] and DUNE [22]. We use the TAUOLA package [42], which decays the tau according to its branching ratios (see Table I) and accounts for tau A. Hadronic Channel polarization effects. We performed our analysis using the Rivet toolkit [43] which is a widely used analysis code for the LHC and As discussed, the dominant background to the hadroni- other high energy physics collider experiments. However, cally decaying taus consists of NC neutrino-argon interac- we find its utility equally applicable to neutrino exper- tions which receive contributions from all three neutrino iments and in this work we premier its use at DUNE. flavors. There are contributions from the NC interactions Tau neutrino interactions typically lead to a high mul- coming from wrong-sign neutrino contaminants but these tiplicity of particles in the final state. This is especially are subdominant as νe, νµ and ντ comprise 0.5%, 3.5% the case for hadronically decaying taus which exhibit sig- and 3.3% of the nominal neutrino beam, respectively, in nificant branching fractions to multiple mesons. In our the far detector. The ντ CC interactions also provide a analysis, we apply a jet clustering algorithm to all visi- small contribution to the signal and this is included in ble final state particles. These include protons, charged the analysis of the hadronic channel. pions, photons (as they lead to electromagnetic show- The first step in constructing the analysis for τhad is ers) and charged kaons but not neutrons or neutrinos. to veto final state particles below the following minimum energy thresholds [14]: 5

± P • π : E > 100 MeV 4. Eother is the total visible energy of the event excluding the leading π−. We veto events with • p: E > 50 MeV P Eother < 600 MeV. miss • γ, e, µ: E > 30 MeV 5. pT is the missing transverse momentum. We veto events with pmiss > 1 GeV. The Monte-Carlo events contain many neutral pions, T which have a decay length cτ ≈ 46.5 µm. As the decay 6. Njet is the number of jets in the final state. We length is much larger than typical nuclear radii, GiBUU veto zero jet events. propagates the neutral pions out of the nucleus and does not decay them. On the other hand, this decay length is We note that the cuts are applied in this order. We also too small to be resolved by DUNE and thus the π0s decay consider the possibility that charge identification is not promptly in the detector. We decay the π0s by boosting possible and alter the analysis such that π− is replaced them to their rest frame then decaying them isotropically with π±. to two photons and boosting the system back to the lab For the signal, we simulated the tau decaying to all frame. For the τhad channel we have analyzed the Monte- possible final states. Therefore, approximately 35% of Carlo output (as generated by GiBUU and TAUOLA) in the events contain electrons and muons and the first cut terms of average multiplicity and energy sum as shown removes this leptonic contribution. To construct the re- in Fig. 3. The left (right) plot shows the average mul- maining cuts we considered that the dominant tau de- tiplicity (energy sum) of visible final state particles as a cays (constituting ∼ 70% of the hadronic channel) are − − − − 0 − − 0 0 function of the true neutrino energy. The events which τ → π ντ , τ → π π ντ and τ → π π π ντ as fill these histograms have had the aforementioned thresh- shown in Table I and the signal tends to contain a hard olds applied to them. The energy threshold for tau pro- π− which motivates the second and third cuts. The nor- duction is evident from the signal (in both beam modes) malized distributions of the kinematic variables of the which activates around Eν ∼ 3 GeV. Moreover the ini- signal (red) and electron, muon and tau (blue, green and tial multiplicity, at low values of Eν , of the signal is ∼ 4 orange respectively) backgrounds are shown in Fig. 4. which corresponds to a few visible hadrons emerging from We note that cuts have not yet been applied to the events the nuclear cascade combined with the tau dominantly that fill these histograms other than (1) and (2). These decaying to two visible pions. Unsurprisingly, the mul- distributions show vital shape information that we used tiplicity of both the signal and background grows as a to design cuts (3)-(6). The upper left plot shows the nor- function of the true neutrino energy and is similar at malized distributions of energies of the leading π− and it high neutrino energies. From the right plot of Fig. 3 we is clear that the signal has a larger proportion of high en- observe that the average visible energy sum of the signal ergy π− while the background has a distinct peak in the displays a threshold while the (dominant) background low energy bins. For the signal, the hardest π− originates can produce low energy and multiplicity final states. We from the tau decay and it carries E & O(100) MeV in en- note that at high values of the true neutrino energy (12- ergy while the background is characterized by many lower 25 GeV) the signal has a larger average visible energy energy hadrons resulting from the hadronization process sum than the background. For this regime, in the case followed by propagation of these hadrons through the of the signal, much of the true neutrino energy will be nucleus via intra-nuclear cascade. A crossover between deposited in the visible final state produced from the tau the signal and background distributions occurs around decays. However, in the case of the background, for the E − ∼ 250 MeV and we place our cut here to enrich πlead same value of true neutrino energy, all the deposited en- the signal and deplete the background. P ergy will result in the hadronic shower which can contain The second observable we consider is Eother and its invisible neutrons. corresponding normalized distribution is shown in the up- In order to optimize the signal-to-background ratio, per right plot of Fig. 4. The shape difference between we study the distributions of several kinematic variables the signal and background is distinct: the background initially assuming charge identification of the pion is pos- is relatively flat in this observable apart from a slight sible. Here we provide a list of these variables and the increase in the distribution around 1 GeV. In contrast, cuts we apply: the signal distribution has a marked dip below 1 GeV. We can observe this dip derives from the tau produc- ± ± 1. Nlep is the number of e and µ . We veto events tion threshold as shown in the right plot of Fig. 3 where containing any such leptons in the final state. the average visible energy has an initial value ∼ 1 GeV for true neutrino energies close to the tau production − − P 2. Nπ is the number of π ’s in the final state. We threshold. For Eother > 1 GeV we observe the signal veto events containing zero π−. increases and this corresponds to the difference in the signal and background spectra as shown in the right plot − 3. πlead is the energy of the leading (highest energy) of Fig. 3. We note that the reason for this difference is π− in each event. We veto events if the leading π− the same in this observable as it was in the aforemen- has E < 250 MeV. tioned histogram. We varied the cut in this observable 6

Highest energy π− All particles but the leading pion 0 14

σ . − lead other d π E d signal dN

∑ 1 1 σ 10 0.12 e d − 1 µ N 0.1 τ

0.08 2 10− 0.06

0.04 signal 3 e 10− 0.02 µ τ 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 1 1 1 10− 10 πlead− [GeV] ∑ Eother [GeV] miss pT Njet 0.45 jet T miss dN dN

0.16 dN

dp signal signal

1 0.4 N 1 N 0.14 e e µ 0.35 µ 0.12 τ 0.3 τ 0.1 0.25 0.08 0.2 0.06 0.15 0.04 0.1 0.02 0.05 0 0 0 0.5 1 1.5 2 2.5 3 0 jet 1 jet 2 jet 3 jet 4 jet 5 jet 6 jet 7 jet 8 jet 9 jet miss pT [GeV] Njet

FIG. 4. The normalized distributions of the hadronic signal for the nominal neutrino flux (define as the CC ντ and ντ contribution) and νe, νe, νµ, νµ and ντ , ντ NC backgrounds as shown in red, blue, green and orange respectively. In these histograms pion charge identification is assumed.

P between 200 ≤ Eother(MeV) ≤ 800 and found that a jet clustering algorithms. In order to be clustered into P veto on events with Eother < 600 MeV reduces the a single jet, visible particles must be within a radius of backgrounds most effectively. R = 0.6, with R = pη2 + φ2 where η = − log tan(θ/2) is The distribution of the missing transverse momentum the pseudo-rapidity (θ is the angle between the particle is shown in the lower left plot of Fig. 4. The invisible and the jet axis) and φ is the azimuthal angle with respect states which contribute to this missing transverse mo- to the jet axis 4. The lower right plot of Fig. 4 shows the mentum vector are neutrinos and neutrons 3. We observe distribution of jet multiplicities. We observe that the that the signal distribution is more strongly peaked than signal has a lower zero jet rate than any of the e, µ or τ the background as the former will always have a ντ in the backgrounds. Moreover, the background peaks at one jet final state. A crossover in the shapes of the signal and while the signal peaks at two jets. We can interpret this miss background occurs at pT ∼ 900 MeV which motivates in the following way: the background predominantly has cut (5). a single jet which emerges from the argon nucleus (see The final cut is on the jet multiplicity and we discuss the lower left image of Fig. 2 where this jet is denoted as this in more detail due to its non-standard application njet). On the other hand, the signal mainly has a two-jet in neutrino physics. Rivet uses FastJet to cluster visible final state where one jet is produced from the tau decay final states into jets. We define the jet to have a minimum (denoted as τjet in the upper left image of Fig. 2) and energy of 1 GeV. In particular, we use the Cambridge- another from the intranuclear cascade. The number of Aachen algorithm [45] which falls into the kT class of

4 We varied the radius in the interval 0.2 ≤ R ≤ 1.0 and found 3 Fermi momentum O(200 MeV) in the argon nucleus may also the analysis was not particularly sensitive to the jet size in this contribute to the missing transverse momentum. range. 7

10 with charge discrimination without charge discrimination 25 τ mode with charge discrimination 8 τ mode without charge discrimination

6 B

√ 15 signal S/ background

4 10

2 5

N miss 1 2 3 4 5 6 7 8 9 10 Nlep π− πlead− Eother pT Njet years

FIG. 5. Cut flow analysis for theP hadronic signal and as- FIG. 6. The solid green (solid purple) shows the signal over sociated NC backgrounds (using the nominal neutrino flux) square root of background as a function of time for the nomi- assuming charge identification of the pion. nal neutrino (tau optimized) mode assuming pion charge identification. The dot-dashed green (dot-dashed purple) shows the signal over square root of background as a function of time for the nominal neutrino (tau optimized) mode without jets in the final state can be as high as nine, although pion charge identification. these higher jet multiplicities are suppressed. This is due to the broad span of the neutrino beam energy. We found the optimal cut, for this definition of a jet, is to veto zero √ fore, the significance decreases to S/ B ∼ 3.1 and a sig- jet events. nificance value of 5 requires approximately 2.5 years of The efficacy of the cuts is summarized in Fig. 5. We data-taking. The improvement in the significance with show the ratio of efficiencies of the signal over the NC charge identification is mainly due to background re- background interactions, as a function of each cut. The duction, as multi-GeV neutral current interactions will application of the cuts to the events should be read from tend to produce comparable amounts of leading π+ and left to right. At this stage, fluxes are not taken into ac- π−, while the signal is dominated by leading π−. Thus, count and this plot simply represents the effectiveness of pion charge identification can be used to further mitigate each cut. The first cut has a value of / ∼ signal background backgrounds without reducing the signal. In the optimal 0.65 which results from removing the leptons from the scenario, a perfect pion charge discrimination would be signal sample. The second cut, which ensures there is at equivalent to an increase of about 17% in exposure in least one negatively charged pion in the final state, dras- tau neutrino analyses. This result demonstrates the non- tically increases the efficiency from ∼ 0.65 to ∼ 4.8 and trivial leverages that LArTPCs may have when making we find this to be the most aggressive requirement on fi- full use of topological information. nal states. The cut on the energy of the leading π− and We applied the same analysis cuts to the tau-optimized high values of missing transverse momenta also have a beam sample and found the significance to be almost significant effect. Further, we find the requirement of at three times higher compared to the nominal beam. The least a single jet in the final state is more effective than solid dark purple shows the significance with charge dis- vetoing P E < 600 MeV. other crimination and after one year of data-taking is ∼ 8.8. Finally, to demonstrate the impact of these series of This corresponds to 433 and 2411 signal and background cuts on ν /ν detection at DUNE, we show the signifi- τ τ √ events respectively. Unsurprisingly, without the charge cance, defined as S/ B where S (B) is the number of discrimination capability the significance is lower with a signal (background) events, as a function of run time in value of ∼ 7.9 which corresponds to 439 and 3077 signal years in Fig. 6. The solid dark green shows the signifi- and background events respectively after the first year of cance for the nominal beam configuration if DUNE has − + running in tau-optimized mode. the√ capability to distinguish π from π and we find S/ B ∼ 3.3 within a year of data-taking. This corresponds to the detection of 79 and 565 signal and background events respectively. In the scenario that charge B. Leptonic Channel identification is not possible, as indicated by dashed light green, the number of signal and background events de- As outlined before, the leptonic decay channels of the tected after one year is 83 and 731 respectively. There- tau are more challenging than their hadronic counter- 8 dN dN

score 0.18 score d signal d 0.25 signal 1 1 N 0.16 background N background

0.14 0.2 0.12 0.15 0.1

0.08 0.1 0.06 0.04 0.05 0.02

0 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 score score

FIG. 7. The normalized distributions of the τe signal (red) FIG. 8. The normalized distributions of the τe signal (red) and associated background (blue) for the nominal neutrino and associated background (blue) for the tau-optimized neu- beam. trino beam. parts. First, the background cross-section (CC interac- The score can be thought of as a new observable that tions from νe and νµ) is larger than the NC background. allows for discrimination of signal and background events. Second, the tau decays to charged leptons at approxi- The kinematic variables or features used to characterize mately half the rate as it does to hadrons. Nonetheless, the signal and background are we pursue the τe channel where the dominant background is the CC interaction of electron neutrinos. We neglect • dφmin is the minimum angle between the leading − the contribution from the νe CC events as the νe com- (highest energy) e and any other visible particle. position of the neutrino beam is approximately 0.5% at − the far detector. We attempted to construct a simple cut • dRmin is the ∆R between leading e and any other and count analysis for τe, in a similar manner to the τhad visible particle. analysis, but we found the significance after one year of data-taking was below 1.0. In light of this, a more effec- • dφmet is the ∆φ between leading lepton and the tive way to discriminate the signal from the background missing transverse momentum vector. is to use a deep neutral net (DNN). In particular, we uti- • E` is the leading lepton energy. lize Keras with tensorflow [46]. Our methodology is as lead follows: miss • ET is the missing transverse energy. 1. Generate signal and background Monte-Carlo samples. • N is the number of visible particles other than leading lepton. 2. Use Rivet, with the same minimum energy thresh- ± olds as before, to calculate kinematic variables or • Nπ± is the number of π ’s. “features” of background and signal events. The • P E is the sum of the energies of all visible signal is assigned a “classification variable” value other particles other than leading lepton. of 1 and the background 0. 3. Separate data sets into training, validation and test • θ` is the angle of the lepton with respect to the samples. beam axis.

4. We train the DNN using the binary cross entropy • Njet is the number of jets where jets have the same loss function with the training data. We use the deﬁnition as in the hadronic channel.

validation data set to guard against overtraining. jet • Elead is the energy of the leading jet. 5. We feed the test data through the trained DNN and for each event it returns a “score” between 0 and We note that the lepton is not included in the jet def- 1. If the event is more background-like its score is inition and we do not apply charge identiﬁcation of the closer to 0 and conversely if it is more signal-like pion in this analysis as the number of negatively and pos- its score is closer to 1. itively charged pions should be approximately the same 9

35 √ nominal beam Mode beam charge id Nsig Nbg S/ B

τ-optimised τhad nominal 79 565 3.3 30 τhad nominal 83 731 3.1

τhad tau-optimized 433 2411 8.8 25 τhad tau-optimized 439 3077 7.9

τe tau-optimized 63 33 11.0 20 B τe nominal 13 32 2.3 √

S/ 15 TABLE II. The number of signal and background events after one year of data-taking in the hadronic (τ ) and leptonic 10 had (τe) channels for nominal and tau-optimized beam modes.

analyses. The backgrounds to τhad comes from NC events 1 2 3 4 5 6 7 8 9 10 which are flavor blind. Compared to the nominal beam years mode, the tau-optimized run presents a higher value of the signal-to-background ratio because there is a larger FIG. 9. The solid purple (dashed pink) shows the signal over fraction of the ντ flux above the tau production thresh- square root of background as a function of time for the τe old. This enriches the signal significantly. On the other channel using the nominal (tau-optmized) neutrino flux. hand, the τe channel receives background contribution mainly from CC νe events. This background is strongly affected by oscillations, as the νe contamination in the for the τe signal and background. Our chosen DNN ar- initial neutrino beam is very small. At high energies, chitecture is a sequential model with a dense input layer ν → ν oscillations are suppressed by 1/E2, see Eq. (1). followed by two hidden layers of depth 100 sandwiching µ e ν Therefore, operating in the tau-optimized mode in the τe a dropout layer with a dropout parameter of value 0.2. case not only significantly enriches the signal, but also The final layer is a sigmoid output layer of depth 1. strongly depletes the background. This effect can be ap- The normalized score distributions of signal and back- preciated in Fig. 1 by comparing the νe (blue) and ντ ground, for nominal neutrino beam mode, are shown in (red) fluxes at DUNE far detector for the nominal (solid) Fig. 7. As expected, the score value of the background and tau-optimized (dashed) runs. peaks at 0 and the signal peaks at 1. We varied where we placed the cut on the score observable and found vetoing events with a score√ less than 0.85 yields the highest significance with S/ B = 2.3 after one year of data- IV. DISCUSSION AND FUTURE PROSPECTS taking corresponding to 13 signal events and 32 background events. The significance as a function of time is The analysis strategy proposed in this paper demon- given by the√ solid dark purple line of Fig. 9 where we strates the vast potential of LArTPC capabilities. There observe a S/ B = 5.0 can be reached after six years of are two clear avenues for further exploration. First, al- running. though we performed an in-depth physics analysis of tau As with the case of the hadronic channel, the de- neutrinos at DUNE, a realistic simulation of detector ef- tectability of tau neutrinos is vastly improved using the fects is required to completely demarcate DUNE’s sen- tau-optimized beam. The normalized score distributions sitivity to beam ντ appearance. We intend to explore, of signal and background are shown in Fig. 8. We found in a forthcoming work, the impact of detector effects on that vetoing score values below 0.8 gave us the optimal the kinematic variables used here and consequently the significance of 11.0 corresponding to 63 signal and 33 signal-to-background ratio. Second, in order to perform background events, respectively, after one year of data- a more robust physics analysis with the tool chain we taking. The significance is indicated by the dashed pink developed here, a detailed understanding of tau neutrino line of Fig. 9. energy reconstruction is needed in addition to the impact Compared to the hadronic tau sample, the τe sensi- of systematic uncertainties on the inferred ντ spectrum. tivity using the nominal beam is smaller but may still Special attention should be given to µ±/π± mis- provide valuable additional information on tau neutri- identification which can enlarge the τhad background. nos. It is remarkable and somewhat surprising that using Moreover, e/γ separation is needed to reject certain NC the tau-optimized beam, the sensitivity of the τe sample events as a background in the τe analysis. This analysis, is comparable to the τhad sensitivity (comparing√ Fig. 6 excluding the missing transverse momentum information, to Fig. 9). The significant enhancement of S/ B for could also be implemented in multi-GeV atmospheric the τe channel in the high energy run can be understood neutrino searches in DUNE and Hyper-Kamiokande [47], qualitatively by comparing the hadronic and leptonic tau with some tuning of the analysis cuts due to much higher 10 √ incoming neutrino energy. achieve S/ B = 5 in just over two years of data-taking Ultimately, several physics analyses could benefit from using only hadronically decaying taus or in six years of applying our tool chain with realistic detector effects and data-taking with leptonically decaying taus. In the tau- a treatment of systematic uncertainties. Amongst those optimized beam mode, after√ a single year of data-taking, are ντ -nucleus interaction measurements [6, 21]; viola- DUNE could achieve S/ B ∼ 9, 11 for the τhad and tions of unitarity in the leptonic mixing matrix [23, 25, τe channels respectively. Moreover, identifying the pion 48–50]; non-standard neutrinos interactions in neutrino charge via pion track topology in DUNE would be equiv- production, detection and propagation [51]; sterile neu- alent to an increase between 17% and 24% in exposure. trino searches [52–55]; and general consistency tests of the three neutrino oscillation paradigm [24, 28]. In our work we demonstrate that the tau-optimized beam sig- ACKNOWLEDGMENTS nificantly enhances the prospects of tau neutrino measurements at DUNE. In particular, running in this mode We thank Kai Gallmeister, Ulrich Mosel and Zbig- improves the detectability of tau neutrinos in the lep- niew Andrzej Was for their invaluable advice on the us- tonic channel. A successful case of a high energy neutrino age of GiBUU and TAUOLA. We are grateful to André beam run can be found in the MINOS experiment history. de Gouvêafor careful reading of the manuscript. We For example, the constraints set by MINOS+, the high thank Kirsty Duffy, Lukas Heinrich, Joshua Isaacson, energy run of MINOS, on eV-scale sterile neutrinos [56] Xiangyang Ju, Kevin Kelly, Claudius Krause, Shirley and large extra dimension models [57] remain leading the Li, Nhan Tranh, Raquel Castillo Fernandez and Tingjun field even in the presence of newer experiments. Yang for helpful discussions on various aspects of this work. This research was supported by the Fermi National Ac- celerator Laboratory (Fermilab), a U.S. Department of V. CONCLUSIONS Energy, Office of Science, HEP User Facility. Fermilab is managed by Fermi Research Alliance, LLC (FRA), acting We have proposed a novel analysis strategy to study under Contract No. DE–AC02–07CH11359. This mate- tau neutrinos with the DUNE experiment. The marriage rial is based upon work supported by the U.S. Depart- of collider tools and neutrino physics allows us to ex- ment of Energy, Office of Science, Office of Advanced Sci- ploit topological features in neutrino events and thereby entific Computing Research, Scientific Discovery through significantly extend DUNE’s physics reach. The Rivet Advanced Computing (SciDAC) program, grant HEP analysis code used for the hadronic channel is available Data Analytics on HPC, No. 1013935. It was supported at [58]. Our results are summarized in Table II and we by the U.S. Department of Energy under contracts DE- find that, in the nominal neutrino beam, DUNE could AC02-76SF00515.

[1] K. Kodama et al. (DONuT), Phys. Rev. D78, 052002 [12] A. Filkins et al. (MINERvA), (2020), arXiv:2002.12496 (2008), arXiv:0711.0728 [hep-ex]. [hep-ex]. [2] N. Agafonova et al. (OPERA), Phys. Rev. [13] S. Aoki et al. (DsTau), JHEP 01, 033 (2020), Lett. 120, 211801 (2018), [Erratum: Phys. Rev. arXiv:1906.03487 [hep-ex]. Lett.121,no.13,139901(2018)], arXiv:1804.04912 [hep-ex]. [14] B. Abi et al. (DUNE), (2020), arXiv:2002.02967 [3] K. Abe et al. (Super-Kamiokande), Phys. Rev. Lett. 110, [physics.ins-det]. 181802 (2013), arXiv:1206.0328 [hep-ex]. [15] C. H. Albright and R. E. Shrock, Phys. Lett. 84B, 123 [4] Z. Li et al. (Super-Kamiokande), Phys. Rev. D98, 052006 (1979). (2018), arXiv:1711.09436 [hep-ex]. [16] P. Astier et al. (NOMAD), Nucl. Phys. B611, 3 (2001), [5] M. G. Aartsen et al. (IceCube), Phys. Rev. D99, 032007 arXiv:hep-ex/0106102 [hep-ex]. (2019), arXiv:1901.05366 [hep-ex]. [17] K. Hagiwara, K. Mawatari, and H. Yokoya, [6] Y. S. Jeong and M. H. Reno, Phys. Rev. D82, 033010 Nucl. Phys. B668, 364 (2003), [Erratum: Nucl. (2010), arXiv:1007.1966 [hep-ph]. Phys.B701,405(2004)], arXiv:hep-ph/0305324 [hep-ph]. [7] J. Blietschau et al. (Gargamelle), Nucl. Phys. B133, 205 [18] J.-M. Levy, J. Phys. G36, 055002 (2009), arXiv:hep- (1978). ph/0407371 [hep-ph]. [8] J. Wolcott et al. (MINERvA), Phys. Rev. Lett. 116, [19] K. Hagiwara, K. Mawatari, and H. Yokoya, Proceedings, 081802 (2016), arXiv:1509.05729 [hep-ex]. 3rd International Workshop on Neutrino-nucleus interac- [9] R. Acciarri et al. (ArgoNeuT), (2020), arXiv:2004.01956 tions in the few GeV region (NUINT 04): Assergi, Italy, [hep-ex]. March 17-21, 2004, Nucl. Phys. Proc. Suppl. 139, 140 [10] K. A. Olive et al. (Particle Data Group), Chin. Phys. (2005), arXiv:hep-ph/0408212 [hep-ph]. C38, 090001 (2014). [20] M. Aoki, K. Hagiwara, K. Mawatari, and H. Yokoya, [11] P. Abratenko et al. (MicroBooNE), (2020), Nucl. Phys. B727, 163 (2005), arXiv:hep-ph/0503050 arXiv:2006.00108 [hep-ex]. [hep-ph]. 11

[21] J. E. Sobczyk, N. Rocco, and J. Nieves, Phys. Rev. arXiv:1311.4496 [hep-ph]. C100, 035501 (2019), arXiv:1906.05656 [nucl-th]. [50] X. Qian, C. Zhang, M. Diwan, and P. Vogel, (2013), [22] J. Conrad, A. de Gouvêa,S. Shalgar, and J. Spitz, Phys. arXiv:1308.5700 [hep-ex]. Rev. D82, 093012 (2010), arXiv:1008.2984 [hep-ph]. [51] P. S. Bhupal Dev et al., SciPost Phys. Proc. 2, 001 (2019), [23] S. Parke and M. Ross-Lonergan, Phys. Rev. D93, 113009 arXiv:1907.00991 [hep-ph]. (2016), arXiv:1508.05095 [hep-ph]. [52] G. H. Collin, C. A. Argelles, J. M. Conrad, and [24] P. B. Denton, (2020), arXiv:2003.04319 [hep-ph]. M. H. Shaevitz, Nucl. Phys. B908, 354 (2016), [25] S. A. R. Ellis, K. J. Kelly, and S. W. Li, (2020), arXiv:1602.00671 [hep-ph]. arXiv:2004.13719 [hep-ph]. [53] S. Gariazzo, C. Giunti, M. Laveder, and Y. F. Li, JHEP [26] A. Rashed and A. Datta, Int. J. Mod. Phys. A32, 06, 135 (2017), arXiv:1703.00860 [hep-ph]. 1750060 (2017), arXiv:1603.09031 [hep-ph]. [54] M. Dentler, . Hernndez-Cabezudo, J. Kopp, P. A. N. [27] D. Meloni, Phys. Lett. B792, 199 (2019), Machado, M. Maltoni, I. Martinez-Soler, and arXiv:1903.08933 [hep-ph]. T. Schwetz, JHEP 08, 010 (2018), arXiv:1803.10661 [hep- [28] A. de Gouvêa, K. J. Kelly, G. V. Stenico, and ph]. P. Pasquini, Phys. Rev. D100, 016004 (2019), [55] A. Ghoshal, A. Giarnetti, and D. Meloni, JHEP 12, 126 arXiv:1904.07265 [hep-ph]. (2019), arXiv:1906.06212 [hep-ph]. [29] F. Kling, (2020), arXiv:2005.03594 [hep-ph]. [56] P. Adamson et al. (MINOS+), Phys. Rev. Lett. 122, [30] “Dune fluxes,” https://home.fnal.gov/~ljf26/ 091803 (2019), arXiv:1710.06488 [hep-ex]. DUNEFluxes/. [57] P. Adamson et al. (MINOS), Phys. Rev. D94, 111101 [31] I. Esteban, M. C. Gonzalez-Garcia, A. Hernandez- (2016), arXiv:1608.06964 [hep-ex]. Cabezudo, M. Maltoni, and T. Schwetz, JHEP 01, 106 [58] P. Machado, H. Schulz, and J. Turner, “Analysis code (2019), arXiv:1811.05487 [hep-ph]. supplement to ”Tau neutrinos at DUNE: new strategies, [32] B. Abi et al. (DUNE), (2020), arXiv:2002.03005 [hep-ex]. new opportunities”,” (2020), available at https://doi. [33] A. Friedland and S. W. Li, Phys. Rev. D99, 036009 org/10.5281/zenodo.3924078, version 1.0. (2019), arXiv:1811.06159 [hep-ph]. [34] I. Nutini, Study of charged particles interaction processes on Ar in the 0.2 - 2.0 GeV energy range through combined information from ionization free charge and scintillation light, Master’s thesis, Florence U. (2015). [35] E. Gramellini, Measurement of the Negative Pion and Positive Kaon Total Hadronic Cross Sections on Argon at the LArIAT Experiment, Ph.D. thesis, Yale U. (2018). [36] T. Suzuki, D. F. Measday, and J. Roalsvig, Phys. Rev. C 35, 2212 (1987). [37] R. Acciarri et al. (MicroBooNE), JINST 12, P09014 (2017), arXiv:1704.02927 [physics.ins-det]. [38] K. Duffy, “Microboone capabilities, measurements and plans,” (2019), nuSTEC Workshop on Neutrino-Nucleus Pion Production in the Resonance Region. [39] G. P. Zeller et al. (NuTeV), Phys. Rev. Lett. 88, 091802 (2002), [Erratum: Phys. Rev. Lett.90,239902(2003)], arXiv:hep-ex/0110059 [hep-ex]. [40] J. A. Formaggio and G. P. Zeller, Rev. Mod. Phys. 84, 1307 (2012), arXiv:1305.7513 [hep-ex]. [41] O. Buss, T. Gaitanos, K. Gallmeister, H. van Hees, M. Kaskulov, O. Lalakulich, A. B. Larionov, T. Leit- ner, J. Weil, and U. Mosel, Phys. Rept. 512, 1 (2012), arXiv:1106.1344 [hep-ph]. [42] S. Jadach, Z. Was, R. Decker, and J. H. Kuhn, Comput. Phys. Commun. 76, 361 (1993). [43] C. Bierlich et al., SciPost Phys. 8, 026 (2020), arXiv:1912.05451 [hep-ph]. [44] M. Cacciari, G. P. Salam, and G. Soyez, Eur. Phys. J. C72, 1896 (2012), arXiv:1111.6097 [hep-ph]. [45] Y. L. Dokshitzer, G. D. Leder, S. Moretti, and B. R. Webber, JHEP 08, 001 (1997), arXiv:hep-ph/9707323 [hep-ph]. [46] F. Chollet et al., “Keras,” https://github.com/ fchollet/keras (2015). [47] K. Abe et al. (Hyper-Kamiokande), (2018), arXiv:1805.04163 [physics.ins-det]. [48] Y. Farzan and A. Yu. Smirnov, Phys. Rev. D65, 113001 (2002), arXiv:hep-ph/0201105 [hep-ph]. [49] H.-J. He and X.-J. Xu, Phys. Rev. D89, 073002 (2014),